Trust and governance · Claims and evidence
Every material statement should reveal what kind of evidence it rests on.
This policy governs the path from source to public claim. It separates observation, attribution, inference, estimation, illustration and proprietary methodology so readers can see what is supported—and what remains uncertain.
Seven public claim states
A label is useful only if it changes how the reader should interpret the statement.
- OBSERVEDObserved
Directly present in the inspected source, file, result or recorded evidence.
- ATTRIBUTEDAttributed
A third-party claim reported with its source and without silently adopting it as SERPIFAI fact.
- INFERREDInferred
A conclusion derived from evidence; the reasoning and material uncertainty should remain visible.
- ESTIMATEDEstimated
A quantified or directional estimate with method, assumptions and limitations stated where material.
- ILLUSTRATIVEIllustrative
A fictional or simplified example used to explain the method, not evidence about a real market or client.
- METHODProprietary method
A SERPIFAI framework, rule or synthesis. External sources may inform it but do not automatically validate it.
- UNRESOLVEDContradicted / unresolved
Material evidence conflicts or is insufficient. The public output should preserve the conflict rather than force certainty.
Interpretation: these are claim categories, not a confidence ranking. The correct state depends on provenance and evidence, not on position in the visual.
Claim state and evidence state are not the same thing
The public claim state explains how a statement is presented. The evidence state records how the underlying support entered the workflow and what can responsibly be concluded from it.
Participant-supplied / self-reported
Provided by a participant, client, operator or stakeholder. Preserve that source role; do not silently relabel the record as independently observed.
Observed
Directly inspected in a named source, file, tool result or recorded event by the reviewer. A separate observed source may corroborate a participant-supplied record without changing that record's origin.
Inferred
A conclusion derived from one or more evidence records. The reasoning and material uncertainty remain part of the evidence state.
Unresolved
Evidence is missing, conflicting or insufficient for the intended use. The gap remains visible rather than being converted into a neutral or reassuring answer.
Evidence strength does not rewrite evidence origin
These evidence-handling states do not replace the seven public claim states above. A stronger overall evidence position can justify a stronger conclusion, but each record keeps the participant-supplied, observed, inferred or unresolved state that describes how it was actually established.
The publication path
Every decision-bearing claim should be able to travel through this sequence.
- 01Classify the statement
Is it fact, source attribution, inference, estimate, illustration or proprietary method?
- 02Record provenance
Identify the source, date/context and the evidence that supports the statement.
- 03Check evidence fit
Ask whether the source actually supports this exact claim, scope and level of certainty.
- 04Search for contradiction
Material disconfirming evidence is not a footnote; it can change or block the claim.
- 05Qualify the wording
Match the language to the evidence. Avoid turning a limited observation into a universal conclusion.
- 06Release or refuse
Publish the claim with its boundary—or withhold it until the evidence supports the intended use.
Good claim discipline vs confidence theatre
| Preferred | Avoid |
|---|---|
| “In the sampled search context, these domains were visible for the query.” | “These are your competitors.” |
| “The source reports X; SERPIFAI uses it here for Y.” | A citation attached to a broader claim the source does not support. |
| “This is an inference from buyer, offer and market evidence.” | An inference written as an observed fact. |
| “The example is fictional and illustrates the method.” | A synthetic example presented as client proof. |
| “Evidence conflicts; the conclusion remains unresolved.” | Averaging contradictory evidence into a reassuring score. |
Evidence quality is multidimensional
A source can be authoritative yet still be a poor fit for the exact claim.
Directness
Does the evidence observe the thing being claimed or only a proxy?
Recency
Could the relevant behavior, product or market have changed since the evidence was recorded?
Independence
Is the source reporting on itself, selling the measured product or independently evaluating it?
Method quality
Can the collection or analysis method support the conclusion?
Context fit
Does geography, buyer, time period, device, market or decision context match?
Decision consequence
How much would the claim matter if it were wrong? Higher consequence demands stronger evidence discipline.
AI assistance does not remove release responsibility
Generative AI may help with research organization, synthesis, drafting or code. It does not convert unsupported text into evidence.
Structure and synthesis
AI can accelerate source discovery, comparison, drafting and transformation when the final material is checked against the relevant evidence and product requirements.
Material publication authority
Material facts, claims, evidence boundaries and consequential methodology changes require explicit review. Human release responsibility remains separate from model output.
When a citation is not enough
Does a citation make a claim true?
No. The source must support the exact claim being made, at the appropriate scope and certainty.
Is participant-supplied evidence the same as observed evidence?
No. Participant-supplied or self-reported evidence keeps that provenance. If a separate source independently corroborates it, record the corroborating observation separately rather than rewriting the original evidence state.
Can a vendor source be used?
Yes, especially for current product behavior or the vendor's own definitions. It should not be treated as independent proof merely because it is detailed.
How should conflicting sources be handled?
Preserve the contradiction when it is material. Investigate provenance, context and method before selecting or combining evidence.
Should every sentence have a source?
No. Source density is not the goal. Material external facts need appropriate evidence; clearly identified SERPIFAI method, explanation and inference should not be disguised as external authority.
The evidence tier sets the claim ceiling
More pages can improve public evidence coverage, but a public-web scan does not become customer or independent evidence by accumulation.
Use public pages to support what a source displays, claims or makes observable in the stated context. Preserve source role and observation date.
CRM, sales conversations, customer feedback and win/loss can support actual comparison behavior when collected and governed appropriately.
Consequential decisions may require a separately accountable verification step rather than stronger wording around the same underlying signal.
Competitive research needs source-role labels
| Source | Strongest normal use | Common overclaim to refuse |
|---|---|---|
| Competitor website | Observed offer, positioning, target language, published market | “Customers actually prefer this” |
| SEO / visibility tool | Estimated rankings, keywords, links, visibility and discovery candidates | “This proves commercial rivalry or actual revenue” |
| Review / community signal | Candidate pains, objections and experience hypotheses | “This sample represents the whole market” |
| CRM / sales / win-loss | Observed buyer comparisons and deal context inside the available dataset | “This automatically generalizes beyond the recorded segment” |
| Independent verification | Higher-assurance review of material claims and scope | “Independent means universally true” |
Evidence and source roles
Google guidance informs people-first content and responsible use of generative AI in publishing; evidence-management and publication-ethics references support provenance and correction discipline. They do not certify SERPIFAI methodology.
- Google Search Central — Creating helpful, reliable, people-first content
- Google Search Central — Optimizing for generative AI features
- CEBMa — Center for Evidence-Based Management
- COPE — Visibility of corrections and retractions
- Klue — Competitive enablement using CRM/conversation evidence
- Crayon — Win/loss analysis
- Product Marketing Alliance — Win/loss analysis
Use the editorial standard for release authority
This policy explains the claim layer. The Editorial and Review Standard explains how an entire page or product reaches release.
Continue to Editorial Review
Review scope, research saturation, contradictions, reader usefulness, technical integrity and reopening triggers together.